Experiments to Bridge the Gap

[Negotiableterms]

The sole purpose of this thread is to design experiments or processes intended to measure the things that people hear, but which don't show up in the current standard measurements. The goal is to bridge the gap between subjectivists and objectivists by changing the things measured and more clearly defining what "hearing" means in this context.

This thread is NOT open for the usual debate, and all posts containing debate will be deleted in their entirety. This means that if you post three pages of good stuff followed by one sentence of debate language, kiss the three pages good-bye, and maybe your TOL privileges as well.

Some generalized examples:

Out of a group of listeners, when cables are changed, how many hear a difference vs how many don't vs how many can't tell. What are the hearing characteristics of the three groups, as tested by audiometer, etc.

Is there a way of testing the quality of brain processing of our hearing? Can we train ourselves to discern more detail? Can this improvement be measured?

Can we predict the sonic results of changing amplifiers based on their impedance curves against the particular speaker used?

Do cables have significant impedance curves, in the setting of a particular amp and speaker combination?

As you can see, we're approaching it from both sides. I hope to wind up with findings such as human hearing is a great deal more discerning than is currently believed, and changes of very small impedance (or other) differences can make audible changes in the sound given the right amp/cable/speaker combination, because of critical interactions in their impedance curves.

Not being a tech expert, I'm making this up...so come on along and help us get it right. All positive attitudes welcome!

 

[crackerkorean]

Is there a way of testing the quality of brain processing of our hearing? Can we train ourselves to discern more detail? Can this improvement be measured?

Can we predict the sonic results of changing amplifiers based on their impedance curves against the particular speaker used?

Do cables have significant impedance curves, in the setting of a particular amp and speaker combination?

I was actually going to do some cable measurements for fun, yea I know im a riot! No matter what the results are I am sure some peoples minds would not be changed but would still be interesting to see.

Testing More detail - I would imagine one way to do this would be to insert either extra noise or sounds into certain parts of a passage of music to see if its audible to the listener. But these sounds would be of a lower magnitude than the surrounding music. But do not let the listener know when the sounds occur. Limiting these occurrences to X number and only let the listener identify that many occurrences. Given enough people being tested we should be able to see if there is a trend.

Granted with this method there is the variable of gear freq range of hearing of the listener.

Testing Cables - You would need an audio oscillator and a known load and take AC voltages at each end from 20Hz to 20KHz. With a sine wave and not a square wave. Many many data points will need to be taken to get an accurate freq response curve.

Just my .02

 

[dumptruck]

Thank you for trying to start a more useful version of the mess I made a few days ago. I would like to point out that a double-blind (or even single-blind) listening test is the easiest way to find out if there is something that should be measured in the first place. I would be happy to contribute to doing this properly in any way that I can.
EDIT: let me revise that to "something that could be especially important to measure"

 

[Andyman]

Is there a way of testing the quality of brain processing of our hearing?

I would think not because I'm not certain on how one would draw objective conclusions from a basically subjective determination.

I would expect one might be able to measure certain objective parameters vis' a vis' neurological processing and compare relative strengths, but to attribute "quality" to them would prove difficult.

An analogy might be to take various chemical compositional measurements of say various single malt Scotch whiskeys and try to determine empirically which has the most quality. You certainly would get data, but I doubt you'd find correlation to "quality". And it certainly would have to be done double blind.

Quality is very tricky business. IIRC, a discussion of what constitutes quality was a major part of "Zen and the Art of Motorcycle Mechanics"

 

[crackerkorean]

Just need to come up with a way that is agreed upon by more than one person. I was speaking with a friend of mine that thought about a good DBX method of testing. The thing is it requires relays to switch between two cables but some complained that the relays would change the sound.

My biggest issue with this complaint is that the relays are a controlled part of the experiment so they exist in listening impressions with both cables. But the listening parties would have to agree that the relays being present wouldn't bias them.

They need to be picked at random and used or even changed at random to see

I would like to point out that a double-blind (or even single-blind) listening test is the easiest way to find out if there is something that should be measured in the first place.

But this is like me saying that the only way to know if there is a difference is to measure it with test equipment. Testing in a controlled environment would tell you if there is any change in the cable.

Listening tests need to be done without the listener knowing what cables are being played so that the knowledge of what cables are being played is not an issue with the experiment. It also needs to be done across a large sampling of people that do and do not have a bias on cables in general.

 

[shimniok]

Might be helpful to focus not as much on quality or correlation as much as being able to discern differences. E.g., don't look for the measurement of the better cable. Look for two cables that are consistently heard to be different, and attempt to measure for the difference. It'd probably help to start with obvious, known measurable difference, and work up towards the really subtle differences.

I don't know that we'd be able to come up with new measurements that loads of brainy people in audio haven't yet... but... it'd be fun trying or at least talking about.

Michael

 

[Andyman]

I don't know if that's a valid way to go.
A blue shirt is different than a red shirt, but how do you determine which is (and which SHOULD BE) preferred???

It's subjective.:yes:

 

[dumptruck]

But this is like me saying that the only way to know if there is a difference is to measure it with test equipment.

Sorry, I don't get what you mean there.

The other things you are describing are part of the definition of blind testing, so I agree. However, a proper blind test can be done on a single person. Your sample is then the number of times you run the test, and your result is evidence of whether that single person can discern a preference between A and B, (or some variation on that). I think this is a more crowd-pleasing way to go, because you don't need to argue about the nature of a sample that is a group of people. It leaves room for the 1/10000 golden-eared person to show their result.

As far as methodology, I would enjoy trying to come up with a repeatable blind test that people can agree on, if anyone is interested.

 

[dumptruck]

P.S. I would much prefer to test things where it is questionable whether a difference can even be noticed. As everyone is already pointing out, testing which of two things is "better" is much more difficult and, in my opinion, less interesting.

 

[dumptruck]

I don't know that we'd be able to come up with new measurements that loads of brainy people in audio haven't yet... but... it'd be fun trying or at least talking about.

Sorry to hog the thread but I missed that remark at first. You are probably right, except for proper blind listening tests. It's practically never done.

 

[crackerkorean]

Sorry, I don't get what you mean there.

The other things you are describing are part of the definition of blind testing, so I agree. However, a proper blind test can be done on a single person. Your sample is then the number of times you run the test, and your result is evidence of whether that single person can discern a preference between A and B, (or some variation on that). I think this is a more crowd-pleasing way to go, because you don't need to argue about the nature of a sample that is a group of people. It leaves room for the 1/10000 golden-eared person to show their result.

As far as methodology, I would enjoy trying to come up with a repeatable blind test that people can agree on, if anyone is interested.

I was wondering why a double blind or single blind test would be the determinate if a cable needs to be tested with test equipment. It could be the other way around, what if there was a resonace on the cable at lets say 10Khz. If change in impedance exists then it should be put to listening tests.

I think regardless of how it sounds or measures, both need to be done and then see if there is a correlation of the results.

I personally dont care about how it sounds just how it is different from a control.

But to bet on one person for their changes when it comes to a cable or cables is a bit thin on the supporting data. If they take a large sample group (>100), and the results are not significantly separate from 50% it would render that test inconclusive.

By getting a large sample of people who are on this site you would approach a more reasonable average across a sample of people who care about audio. What if that one person is the person that cannot tell the difference between a boom box and the perfect system.

Unless I completely misunderstand and you mean one person at a time which i can go with as long as many people are sampled and used in final data analysis.

Repeatable results are key when it comes to comparisons and proper data collection.

 

[Negotiableterms]

Whoah!!!!

Let's leave off the whole discussion of blind listening tests until we figure out exactly what we're going to have the poor folk listen to.

Here's an example of the kind of thinking I'm looking for:

We play a track of computer generated pink noise. Then a second track, in which we've had the computer reduce the so-called "warmth region" by 0.5dB.

My theory is that a fair % of people will be able to hear the difference, because it will occur over a range of frequencies, instead of just one, where the definition says 1dB is the minimum.

If that theory proves out, we try 0.25dB and see what happens. Then, instead of pink noise, we try white, chords, etc.

The goal is to see whether the processing going on in our brains effects the basic concept of the dB.

 

[luvvinvinyl]

With regard to sampling, at last year's Fest, Dr. Joe D'Appolito was asked about voicing speakers, and subjective opinion. "How can you voice a speaker to be marketed successfully?" He went on at length about how their collected data, sampling of people from all walks of life, showed them a range of opinions about speaker voicing. Their research led them to a data set that quantifies people's opinions about the sound of speakers. They use that data, and try to voice their speakers to satisfy the data points for that research. He went on to speak about the fact that they cannot please everybody, but they try to get the best fit the can, compared to the upper range of their data.

I found it fascinating that they attempt to quantify opinion, and base their designs on that. Final voicing approval came after listening sessions from many people in the plant. The Snells have a 'family voice', and are being well-received in the industry, moreso with the newest models.

I rambled on about this for the purpose of demonstrating that it is possible to creat a workable model of opinions about something as subjective as "Does this speaker sound good to you?" Dr. Joe described it as a monumental task, but, with enough time, money and effort, it can be done well enough to be of value as a design parameter.

Is that the direction in which you hope to be headed?

 

[luvvinvinyl]

Crossposted

 

[Andyman]

I'm not sure what the OP is searching for here. It seems he's looking for a methodology that may determine if there are measurements that may predict a priori how a cable may sound, or even more broadly, a set of measurements that will indicate if cables DO in fact sound different.

So perhaps it's not so much about which is better, but if there is a difference at all?
And that would almost require repeated exposure to the same conditions with repeated changes (and lack of changes) without the subject being made aware of the fact that anything was changing so they don't subconsciously go "looking" for change and notice it on their own.

That would be extremely difficult as it would require an almost unconscious critical listening mode. Or truly, sledgehammer night and day differences readily discernible.

No wonder there haven't been many landmark studies here...

 

[luvvinvinyl]

Ok, so you are really talking about the objective side of hearing. Would you set up a correlation between audiometric testing, i.e. a person's threshhold ability to hear a frequency or frequencies, and their perception results?

If you don't set up an individual baseline for each test subject, can you draw conclusions about their test results? My experience with audiometric testing, every 3 years, for my license, is that each frequency range has an associated threshhold for that range. The graph looks very much like a frequency response graph for a particular speaker. If test subjects have a particular insensitivity in a range of frequencies. that would affect their ability to judge changes in that range.

Just freewheelin' here.

 

[Andyman]

With regard to sampling, at last year's Fest, Dr. Joe D'Appolito was asked about voicing speakers, and subjective opinion. "How can you voice a speaker to be marketed successfully?" He went on at length about how their collected data, sampling of people from all walks of life, showed them a range of opinions about speaker voicing. Their research led them to a data set that quantifies people's opinions about the sound of speakers. They use that data, and try to voice their speakers to satisfy the data points for that research. He went on to speak about the fact that they cannot please everybody, but they try to get the best fit the can, compared to the upper range of their data.

I found it fascinating that they attempt to quantify opinion, and base their designs on that. Final voicing approval came after listening sessions from many people in the plant. The Snells have a 'family voice', and are being well-received in the industry, moreso with the newest models.

I rambled on about this for the purpose of demonstrating that it is possible to creat a workable model of opinions about something as subjective as "Does this speaker sound good to you?" Dr. Joe described it as a monumental task, but, with enough time, money and effort, it can be done well enough to be of value as a design parameter.

That's very interesting and sensible.

FWIW, I bet that's exactly what Bose does, with a little bit marketing tossed in to "tell" buyers what they want to hear.

It's all a matter of defining one's target market; just like when music was mixed to sound good on car radios.

Now back to out regularly scheduled programming

 

[dumptruck]

Unless I completely misunderstand and you mean one person at a time which i can go with as long as many people are sampled and used in final data analysis.

Yep, it's like this: If you test a whole ton of people with one (or a few) A/B comparisons, then all you get is an estimate of how many people prefer A, B, or don't know. If you test one person hearing one thing, then the other, repeated many times, then you have very strong evidence of whether that individual can or cannot discern a preference. This is meaningful, and not at all the same as someone doing a test on their own. Then the same could be done with as many people as willing, giving you an increasing picture of the percentage of people that can do it.

I was wondering why a double blind or single blind test would be the determinate if a cable needs to be tested with test equipment. It could be the other way around, what if there was a resonace on the cable at lets say 10Khz. If change in impedance exists then it should be put to listening tests.

Okay now I get it. You're right, both are fine ways to go about it. Which is the better way for any particular case is actually impossible to say unless you already know how one of the two types of tests will come out.

Repeatable results are key when it comes to comparisons and proper data collection.

Amen. Since it will be impossible for any reasonably-priced experiment to replicate all gear, rooms, etc., this is another advantage of doing one person at a time with many many a/b tests for each person. If the methodology stays the same, the gear and room can change without making the results totally useless. Of course, the gear and room would need to be carefully noted, though.

 

[crackerkorean]

I apologize I must emphasize I don't care about if it sounds good but if there is a difference between the test subjects (cables etc).

The Equal Loundness contours and weighted filters that are used in acoustics were calculated across a very large sample of people.

Doing testing with listeners will give a a good idea if there are trends on a difference one way or another. But as pointed out before this is very difficult and tedious

As for what to listen to that would be a difficult one to do. Modifying tones is an easy thing to do and very quantifiable but not to amusing to listen to.

I guess what would be good to start would be to find out what kind of things cables are claimed to do. (lower noise floor, better high end, etc). I think from there test tracks need to be created to test these parameters. If your testing for noise add white noise at -120dB and -115dB etc. But the problem with this is what is the noise floor of the equipment being used also.

It would almost be best to send a test system around, unfortunately this is not very dooable.

Statistics was a blast but real world analysis is a PITA.

Hope my thoughts and posts are not too all over the place.

 

[dumptruck]

These problems are real problems, but not total roadblocks for listening tests done many times on each subject as I've been advocating. For example, (I bet you could guess I'd pick this example) a speaker wire test.

In a speaker wire test each experimenter would test 1 or more subjects, many times each, using a predetermined gauge of lampcord vs any more expensive cable they happened to have. Many things would have to be noted (gear, terminals, music/sounds, and on and on), but if many people did the test the results would be useful despite the mismatched settings.

You might see that no one could tell a difference no matter what was used - that would obviously be interesting. You might see that only a few people could tell a difference, but it was always with certain wires or certain types of music or certain amps or speaker classes, etc. Then you'd want to get more tests with similar gear, and eventually find out what allowed the difference to come through. You might see that almost everyone was hearing differences, proving that speaker wire matters more than many say it does, again, obviously interesting.

If you could get a good number of people to participate (maybe not very likely), something would almost surely pop up worth pursuing further. Exceptions, as well as trends, are important if you test single subjects many times.